No abstract

Managing the dynamic behavioral changes of a production plant process to keep to a production plan is a challenge and requires the ability to predict the dynamic behavior of processes and alter any controls, as needed, to adhere as closely as possible to the plan. This paper presents a novel solution (called Cognitive Plant Advisor) based on the use of advanced machine learning to learn complex dynamics from sensor data coupled with mathematical programming to optimize the operations of a production plant. The Cognitive Plant Advisor provides set point recommedations for a 12-72 hour horizon to (i) improve throughput, or (ii) provide optimal recovery plan for a disruption. This advisory system has the potential to improve throughput by upto 1% of total production.